8/3/2019 ED015_Karimi_EA

1/1

8th Latin American and Caribbean Conference for Engineering and Technology

Arequipa, Per WE1-1 June 1-4, 2010

A Fuzzy Scheduler for Crossbar Switches

Masoumeh KarimiFlorida International University, Miami, mkari001@fiu.edu

INTRODUCTION

Crossbar switches have received significantattention over the past two decades. They

consist of 2N buses to connect N input ports to

N output ports in a matrix manner. Differentarbitration rules can be applied to schedule the

incoming packets in crossbar switches, such asRound Robin (RR), Earliest Deadline First

(EDF), Longest Queue First (LQF), First-In

First-Out (FIFO), and Random (RD). In these

rules, only one criterion is considered to identifyand serve the eligible packets. However, in this

paper, we consider a combined variable todynamically schedule the best effort flows. Inour approach, queues are served based on the

reconfigurable weights. The weight of each

queue is a fuzzy combination of two parameters,length of the queue in the input buffer and the

departure deadline of the packets.

THE STRUCTURE OF THE FUZZY SCHEDULER

As depicted in Figure 1, the scheduler consists

of two inputs, buffer occupancy and departure

deadline of the packets; and one output todetermine the weight of the incoming traffic to

make the scheduling decision.

Figure 1: A Fuzzy Scheduler

There are three membership functions for each

input and four linguistics terms for the output.

The linguistic terms related to the input bufferoccupancy are light, medium, and full

corresponding to the queue length that has

occupied the buffer capacity. The departuredeadline is described by three expressions:

short, medium, and long to represent the

remaining time for the packet expiration. The

four linguistic variables assigned to the output

are low, medium, high, and too high to

determine the weight of each queue for thepacket scheduling. For instance, too high

implies that particular packet has the highest

priority for the scheduling process.

After applying the inputs to the fuzzyscheduler, the inference engine computes the

output corresponding to each rule. A set of If-

Then rule is used to derive a consequencesimilar to the human reasoning process. For

example, If the buffer occupancy is full and the

departure deadline is short, then the weight is

too high. Next, after organizing the fuzzyconditional rules, the inputs are combined based

on the Mamdani model to produce the values forthe output. The standard Centroid method is

applied to calculate a crisp output value.

Accordingly, a weight is allocated to each

packet and then, arrival packets are dynamicallyscheduled through the crossbar switch based on

their weights. The metrics used to evaluate the

performance of the fuzzy scheduler are thethroughput and the average delay.

REFERENCES

Karimi M., Sun Z., Pan D., and Chen Z. (2009).

Packet-mode Asynchronous Scheduling

Algorithm for Partially Buffered Crossbar

Switches, IEEE Global Communications

Conference (Globecom09), Honolulu, HI.

Kurose J. and Ross K. (2007). Computer

networking: a top-down approach, Addison

Wesley, 4th edition.

Ross T. J., (2005). Fuzzy Logic with Engineering

Applications, Wiley, 2nd edition.

Gomathy C. and Shanmugavel S. (2004). Anefficient fuzzy based priority scheduler for

mobile ad hoc networks and performance

analysis for various mobility models, IEEE

Wireless Commun. and Netw. Conf. (WCNC04),

vol. 2, pp. 10871092, Atlanta, Ga, USA.